app/pages.py

import streamlit as st
from app.draw_diagram import *

def dashboard():

    with st.container():
        st.title("SeaEval")
   
        st.markdown("""
            [gh]: https://github.com/SeaEval/SeaEval
            [![GitHub Repo stars](https://img.shields.io/github/stars/SeaEval/SeaEval?style=social)][gh]
            """)


    st.markdown("""
            ### Changelog

            - **Dec 2024**:
                - Updated results for **Cross-MMLU**, **Cross-LogiQA**, **Cross-XQuad**, **MMLU**, **IndoMMLU**, and **SG-Eval-v2** with new prompts (simple prompts to encourage reasoning).
                - Added new models: **SEA-LION v3**, **Gemma-2**, and **Sailor 2**.

            - **Nov 2024**:
                - Updated layout and added support for comparison between models with similar sizes.
            """)

    st.divider()

    st.markdown("#### What is [SeaEval](https://seaeval.github.io/)?")

    with st.container():
        left_co, cent_co,last_co = st.columns(3)
        with cent_co:
            st.image("./style/seaeval_overall.png", 
                    #  caption="SeaEval data range", 
                    width=500)
        st.markdown('''

                    ''')
        st.markdown("##### A benchmark for multilingual, multicultral foundation model evaluation consisting of >30 dataset and we are keep expanding over time.")
        st.markdown(''':star: How models understand and reason with natural language?    
                    :balloon: Languages: English, Chinese, Malay, Spainish, Indonedian, Vietnamese, Filipino.
                    ''')
        
        st.markdown(''':star: How models comprehend cultural practices, nuances and values?    
                    :balloon: 4 new datasets on Cultural Understanding.
                    ''')

        st.markdown(''':star: How models perform across languages in terms of consistency?    
                    :balloon: 2 new datasets with curated metrics for Cross-Linugal Consistency.
                    ''')
    with st.container():
        left_co, cent_co,last_co = st.columns(3)
        with cent_co:
            st.image("./style/consistency.png", 
                    #  caption="SeaEval data range", 
                    width=500)
        st.markdown("##### Evaluation with enhanced cross-lingual capabilities.")
        st.markdown(''':star: How models perform according to different (paraphrased) instructions?    
                    :balloon: Each dataset is equipped with 5 different prompts to avoid randomness introduced by instructions,
                    which is non-negligible..
                    ''')
        
        st.markdown(''':star: Multilingual accuracy and performance consistency across languages.    
                    :balloon: If you can answer the question in your native language, can you answer the same question 
                    correctly in your second/third language?
                    ''')

    st.divider()
    with st.container():
        st.markdown("##### Citations")

        st.markdown('''
                    ```
                        @article{SeaEval,
                        title={SeaEval for Multilingual Foundation Models: From Cross-Lingual Alignment to Cultural Reasoning},
                        author={Wang, Bin and Liu, Zhengyuan and Huang, Xin and Jiao, Fangkai and Ding, Yang and Aw, Ai Ti and Chen, Nancy F.},
                        journal={NAACL},
                        year={2024}
                        }
                    ```
                    ''')
        

def cross_lingual_consistency():
    st.title("Task: Cross-Lingual Consistency")
    
    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = [
                        'Cross-MMLU', 
                        'Cross-XQUAD', 
                        'Cross-LogiQA', 
                        ]
    
    category_one_dict = {
                        'Zero Shot': 'zero_shot',
                        'Few Shot' : 'few_shot'
                         }
    
    category_two_dict = {
                        'Cross-MMLU'            : 'cross_mmlu_no_prompt',
                        'Cross-XQUAD'           : 'cross_xquad_no_prompt',
                        'Cross-LogiQA'          : 'cross_logiqa_no_prompt',
                         }

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])

    with right:
        sort = st.selectbox('Sort (For Chart)', ['Accuracy','Cross-Lingual Consistency', 'AC3',
                                     'English', 'Chinese', 'Spanish', 'Vietnamese'])
       
    sortby = 'Ascending'

    if category_one or category_two or sort or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        
        draw('cross_lingual', category_one, category_two, sort, sortby, model_size_range)


def cultural_reasoning():
    st.title("Task: Cultural Reasoning")

    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = [
                        'SG-EVAL-v2-MCQ',
                        'SG-EVAL-v2-Open-Ended',
                        'SG-EVAL-v1-Cleaned',
                        'SG-EVAL-v1', 
                        'CN-EVAL', 
                        'PH-EVAL', 
                        'US-EVAL'
                        ]
    
    category_one_dict = {'Zero Shot': 'zero_shot', 
                         'Few Shot': 'few_shot'
                         }
    
    category_two_dict = {
                         'SG-EVAL-v2-MCQ'          : 'sg_eval_v2_mcq_no_prompt',
                         'SG-EVAL-v1'              : 'sg_eval',
                         'SG-EVAL-v1-Cleaned'      : 'sg_eval_v1_cleaned',
                         'SG-EVAL-v2-Open-Ended'   : 'sg_eval_v2_open',
                         'US-EVAL'                 : 'us_eval',
                         'CN-EVAL'                 : 'cn_eval',
                         'PH-EVAL'                 : 'ph_eval'
                         }

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])

    sortby = 'Ascending'
    
    if category_one or category_two or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        draw('cultural_reasoning', category_one, category_two, 'Accuracy', sortby, model_size_range)


def general_reasoning():
    st.title("Task: General Reasoning")

    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = [
                        'IndoMMLU',
                        'MMLU', 
                        'CMMLU', 
                        'C-Eval', 
                        'ZBench', 
                        ]
    
    category_one_dict = {
                        'Zero Shot': 'zero_shot',
                        'Few Shot' : 'few_shot'
                         }

    category_two_dict = {
                         'IndoMMLU': 'indommlu_no_prompt',
                         'MMLU'    : 'mmlu_no_prompt',
                         'C-Eval'  : 'c_eval',
                         'CMMLU'   : 'cmmlu',
                         'ZBench'  : 'zbench',
                         }

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])
    
    sortby = 'Ascending'

    if category_one or category_two or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        draw('general_reasoning', category_one, category_two, 'Accuracy', sortby, model_size_range)


def flores():
    st.title("Task: FLORES-Translation")

    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = [
                        'Indonesian to English',
                        'Vitenamese to English', 
                        'Chinese to English', 
                        'Malay to English'
                        ]
    
    category_one_dict = {
                        'Zero Shot': 'zero_shot',
                        'Few Shot' : 'few_shot'
                         }
    category_two_dict = {
                        'Indonesian to English': 'ind2eng',
                        'Vitenamese to English': 'vie2eng',
                        'Chinese to English'   : 'zho2eng',
                        'Malay to English'     : 'zsm2eng'
                         }
    

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])

    sortby = 'Ascending'

    if category_one or category_two or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        draw('flores_translation', category_one, category_two, 'BLEU', sortby, model_size_range)


def emotion():
    st.title("Task: Emotion")

    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = [
                        'Indonesian Emotion  Classification', 
                        'SST2',
                        ]
    
    category_one_dict = {'Zero Shot': 'zero_shot', 
                         'Few Shot': 'few_shot'}
    category_two_dict = {
                            'Indonesian Emotion  Classification': 'ind_emotion',
                            'SST2'                              : 'sst2'
                             }

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])
    
    sortby = 'Ascending'


    if category_one or category_two or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        draw('emotion', category_one, category_two, 'Accuracy', sortby, model_size_range)

def dialogue():
    st.title("Task: Dialogue")

    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = [
                        'DREAM', 
                        'SAMSum', 
                        'DialogSum',
                        ]
    
    category_one_dict = {
                        'Zero Shot': 'zero_shot',
                        'Few Shot' : 'few_shot'
                         }
    category_two_dict = {
                        'DREAM'    : 'dream',
                        'SAMSum'   : 'samsum',
                        'DialogSum': 'dialogsum'
                         }

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with right:
        if category_two == 'DREAM':
            sort = st.selectbox('Sort', ['Accuracy'])
        else:
            sort = st.selectbox('Sort', ['Average', 'ROUGE-1', 'ROUGE-2', 'ROUGE-L'])

    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])
    
    sortby = 'Ascending'

    if category_one or category_two or sort or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        draw('dialogue', category_one, category_two, sort, sortby, model_size_range)


def fundamental_nlp_tasks():
    st.title("Task: Fundamental NLP Tasks")

    filters_levelone = ['Zero Shot', 'Few Shot']
    filters_leveltwo = ['OCNLI', 'C3', 'COLA', 'QQP', 'MNLI', 'QNLI', 'WNLI', 'RTE', 'MRPC']
    
    category_one_dict = {
                         'Zero Shot': 'zero_shot',
                         'Few Shot' : 'few_shot'
                         }
    category_two_dict = {
                        'OCNLI': 'ocnli',
                        'C3'   : 'c3',
                        'COLA' : 'cola',
                        'QQP'  : 'qqp',
                        'MNLI' : 'mnli',
                        'QNLI' : 'qnli',
                        'WNLI' : 'wnli',
                        'RTE'  : 'rte',
                        'MRPC' : 'mrpc'
                        }

    left, center, middle, _, right = st.columns([0.2, 0.2, 0.2, 0.2 ,0.2])
    with left:
        category_one = st.selectbox('Zero or Few Shot', filters_levelone)
    with center:
        category_two = st.selectbox('Dataset', filters_leveltwo)
    with middle:
        model_size_range = st.selectbox('Model Size', ['All', '<10B', '10B-30B', '>30B'])
    
    sortby = 'Ascending'

    if category_one or category_two or sortby:
        category_one = category_one_dict[category_one]
        category_two = category_two_dict[category_two]
        draw('fundamental_nlp_tasks', category_one, category_two, 'Accuracy', sortby, model_size_range)